Research map / 3D vision / Depth and stereo
Trap Attention: Monocular Depth Estimation With Manual Traps
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add trap attention (TA) to the trap-network decoder (ablation) | Trap-S · NYU Depth-v2 | delta<1.25: +1.0 points (0.883 to 0.892 for S+TA) | Helped |
| add trap interpolation (TI) and trap attention with 4-block block selection (ablation) | Trap-S · NYU Depth-v2 | Abs Rel: 0.120 to 0.100 | Helped |
| remove manual traps from Trap-M (ablation) | Trap-M · NYU Depth-v2 | Abs Rel: 0.092 to 0.112 | Hurt |
| remove rounding operation from trap functions (ablation) | Trap-M · NYU Depth-v2 | delta<1.25: 0.925 to 0.920 | Hurt |
| replace t3 with |sin(x^2)| (ablation) | Trap-M · NYU Depth-v2 | delta<1.25: 0.925 to 0.923 | Hurt |
| replace trap functions with ReLU, GELU, SiLU, or 4 learning weights (ablation) | Trap-M · NYU Depth-v2 | Abs Rel: 0.092 vs 0.111 (ReLU), 0.099 (GELU), 0.111 (SiLU), 0.109 (4 learning weights) | Hurt |
| remove block selection from Trap-M (M+TI+TA without BS) (ablation) | Trap-M · NYU Depth-v2 | Abs Rel: 0.099 to 0.092 | Hurt |
| trap interpolation replacing bilinear interpolation (ablation) | decoder upsampling · NYU Depth-v2 | Abs Rel: reduces Abs Rel by 6.1% and RMSE by 2.2% | Helped |
| Trap-S with about 36% of parameters vs. prior SOTA small models | Trap-S vs BTS/AdaBin · NYU Depth-v2 | delta<1.25: 0.910 vs AdaBin 0.903 | Helped |
| Trap-L (Swin-L encoder) vs NeW CRFs | Trap-L · KITTI Eigen split | Abs Rel: 0.050 vs 0.052 | Helped |
| Trap-L vs NeW CRFs cross-dataset generalization | Trap-L · SUN RGB-D (NYU-pretrained) | Abs Rel: reduces Abs Rel by 11.3% vs prior SOTA | Helped |
| increase number of block-selection candidate blocks from 2 to 8 (Trap-M) (ablation) | Trap-M · NYU Depth-v2 | delta<1.25: 0.916 (none) -> 0.917 (BS2) -> 0.917 (BS4) -> 0.920 (BS6) -> 0.925 (BS8) | Mixed |
Closest papers by meaning
- Patch-Wise Attention Network for Monocular Depth Estimation (AAAI 2021 · similarity 0.90)
- Depth Prompting for Sensor-Agnostic Depth Estimation (CVPR 2024 · similarity 0.86)
- Depth Pro: Sharp Monocular Metric Depth in Less Than a Second (ICLR 2025 · similarity 0.85)
- Depth Anything: Unleashing the Power of Large-Scale Unlabeled Data (CVPR 2024 · similarity 0.85)
- DepthCues: Evaluating Monocular Depth Perception in Large Vision Models (CVPR 2025 · similarity 0.85)
- DME: Unveiling the Bias for Better Generalized Monocular Depth Estimation (AAAI 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.